Fixing PBC conv_args (draft)

[RylieWeaver]

This PR aims to address the issue related to generating conv_args that are in-line with periodic boundary conditions.

The following implementation should be robust under the following assumptions, which I think should always be true:
(1) radius < min(supercell_size) / 2
(2) The edge_index has been generated correctly with periodic boundary conditions
(3) data.supercell_size exists if and only if pbc conditions are used

Currently, only two models are updated to use the new functionality. If this is a good solution, it can be easily extended to the others.

[RylieWeaver]

@allaffa This should give an idea of the implementation I've been working with. I can still take a further look at the Facebook implementation you mentioned, but since this one is already working and I'm familiar with I wanted to show you what it would look like first. Let me know your thoughts.